Completely refactor the LLM integration method to easily replace the LLM used by each module and optimize the retransmission mechanism.

QueryEngine/nodes/summary_node.py

@@ -98,11 +98,10 @@ class FirstSummaryNode(StateMutationNode):
             
             self.log_info("正在生成首次段落总结")
             
-            # 调用LLM，增加max_tokens以支持更长的总结
+            # 调用LLM生成总结
             response = self.llm_client.invoke(
-                SYSTEM_PROMPT_FIRST_SUMMARY, 
+                SYSTEM_PROMPT_FIRST_SUMMARY,
                 message,
-                max_tokens=8192  # 支持更长的总结内容
             )
             
             # 处理响应
@@ -267,11 +266,10 @@ class ReflectionSummaryNode(StateMutationNode):
             
             self.log_info("正在生成反思总结")
             
-            # 调用LLM，增加max_tokens以支持更长的总结
+            # 调用LLM生成总结
             response = self.llm_client.invoke(
-                SYSTEM_PROMPT_REFLECTION_SUMMARY, 
+                SYSTEM_PROMPT_REFLECTION_SUMMARY,
                 message,
-                max_tokens=8192  # 支持更长的总结内容
             )
             
             # 处理响应